A programmable logic device includes a plurality of logic blocks. Each logic block includes a plurality of macrocells, with each macrocell being configurable to register a sum of product term output. In addition, the macrocells within each logic block are arranged from a first macrocell to a last macrocell. Each macrocell is associated with a carry-in and a carry-out signal. The macrocells are configured to support a carry cascade such that the carry-out signal from the first macrocell becomes the carry-in signal for the second macrocell, and so on.
|
8. A programmable logic device including a plurality of logic blocks, each logic block comprising:
a plurality of product term circuits each operable to provide a product term output;
a plurality of n or gates, wherein each or gate is configured to receive a subset of the product term outputs and provide a sum of products output;
a plurality of n macrocells corresponding to the plurality of or gates, wherein each macrocell is configurable to register the corresponding sum of products output and wherein the macrocells are arranged from a first macrocell to an nth macrocell; and
a means for generating a carry cascade from the first macrocell to the nth macrocell, wherein, for each macrocell, the means uses the sum of product term output from the corresponding or gate to control the carry propagation.
14. A method, comprising:
providing a logic block having a plurality of macrocells arranged from a first macrocell to an nth macrocell, each macrocell being configurable to register a corresponding sum of products output;
generating a carry-in signal for the first macrocell;
generating a carry-out signal from either the carry-in signal or a product term output, wherein selection of either the carry-in signal or the product term output is based upon the sum of products output corresponding to the first macrocell; and
continuing the carry cascade such that the carry-out signal from the first macrocell becomes a carry-in signal for the second macrocell, a carry-out signal for the second macrocell becomes the carry-in signal for the third macrocell, and so on, wherein the carry generation at each macrocell is based upon its corresponding sum of products output.
18. A programmable logic device including a plurality of logic blocks, one or more logic blocks comprising:
a plurality of product term circuits each operable to provide a product term output;
a plurality of or gates each configured to receive a subset of the product term outputs and provide a sum of products output;
a plurality of macrocells each configurable to register the sum of products output from an or gate; and
a plurality of multiplexers corresponding to the plurality of macrocells, each multiplexer configurable to select between a product term output and a carry-in signal to provide a carry-out signal;
wherein the multiplexers are arranged to form a carry cascade such that the carryout signal from a first multiplexer becomes the carry-in signal for a second multiplexer, and the carry-out signal from the second multiplexer becomes the carry-in signal for a third multiplexer.
1. A programmable logic device including a plurality of logic blocks, each logic block, comprising:
a plurality of product term circuits each operable to provide a product term output;
a plurality of m or gates, wherein each or gate is configured to receive a subset of the product term outputs and provide a sum of products output;
a plurality of m macrocells corresponding to the plurality of m or gates, wherein each macrocell is configurable to register the sum of products output from its corresponding or gate; and
a plurality of m multiplexers corresponding to the plurality of m or gates and the plurality of m macrocells, wherein each multiplexer is configured to select between a product term output and a carry-in signal to provide a carry-out signal, and wherein the multiplexers are arranged from a first multiplexer to an mth multiplexer to form a carry cascade such that the carry-out signal from the first multiplexer becomes the carry-in signal for the second multiplexer, the carry-out signal from the second multiplexer becomes the carry-in signal for the third multiplexer, and so on.
2. The programmable logic device of
3. The programmable logic device of
a plurality of m exclusive or gates corresponding to the plurality of m or gates, wherein each exclusive or gate is configured to receive the sum of product output from its corresponding or gate and the carry-in signal to generate a sum signal.
4. The programmable logic device of
5. The programmable logic device of
6. The programmable logic device of
7. The programmable logic device of
9. The programmable logic device of
10. The programmable logic device of
11. The programmable logic device of
12. The programmable logic device of
13. The programmable logic device of
15. The method of
for each macrocell, generating a sum signal using the macrocell's carry-in signal and the corresponding sum of product output.
17. The method of
for a given logic block, receiving the carry-out signal from the nth macrocell, and
routing the received carry-out signal into the carry-in signal for the first macrocell in another logic block.
19. The programmable logic device of
a plurality of exclusive or gates each configured to receive the sum of product output from an or gate and the carry-in signal to generate a sum signal.
20. The programmable logic device of
|
This patent application is a continuation of U.S. Ser. No. 10/133,106, filed Apr. 26, 2002, issued as U.S. Pat. No. 6,765,408, which in turn claims the benefit of U.S. Provisional Patent Application No. 60/356,507, filed on Feb. 11, 2002, both of which applications are hereby incorporated by reference in their entirety.
The following co-assigned U.S. patent applications and Patents are hereby incorporated by reference in their entireties: (1) U.S. patent application Ser. No. 09/704,487, entitled “Wide Input Programmable Logic System And Method” filed on Nov. 2, 2000, issued as U.S. Pat. No. 6,507,212; and (2) U.S. Pat. No. 6,067,252, entitled “Electrically Erasable Non-Volatile Memory Cell With Virtually No Power Dissipation”, filed on May 26, 1999.
The present invention relates generally to programmable devices. Specifically, the present invention relates to logic blocks having macrocells supporting a carry cascade.
Field-programmable gate arrays (FPGAs) and programmable logic devices (PLDs) have been used in data communication and telecommunication systems. Conventional PLDs and FPGAs have a predetermined number of non-configurable blocks, where each block performs a pre-determined function. Some currently-available CPLD products comprise arrays of logic cells. On high-density CPLD products, memory modules may be included. The memory module is usually limited to SRAM.
Conventional PLD devices and methods have several drawbacks, such as footprint requirements on a circuit board, limited speed and limited data processing capabilities.
One aspect of the invention relates to a programmable logic device including a plurality of logic blocks. Each logic block includes a plurality of product term circuits each operable to provide a product term output. A plurality of M OR gates are configured to receive the product term outputs, wherein each OR gate is configured to receive a subset of the product term outputs and provide a sum of products output. A plurality of M macrocells corresponds to the plurality of M OR gates, wherein each macrocell is configurable to register the sum of products output from its corresponding OR gate. A plurality of M multiplexers corresponds to the plurality of M OR gates and the plurality of M macrocells, wherein each multiplexer is configured to select between a product term output and a carry-in signal to provide a carry-out signal, and wherein the multiplexers are arranged from a first multiplexer to an Mth multiplexer to form a carry cascade such that the carry-out signal from the first multiplexer becomes the carry-in signal for the second multiplexer, the carry-out signal from the second multiplexer becomes the carry-in signal for the third multiplexer, and so on.
Another aspect of the invention relates to a method including the steps of: (a) providing a logic block having a plurality of macrocells arranged from a first macrocell to an Nth macrocell, each macrocell being configurable to register a corresponding sum of products output; (b) generating a carry-in signal for the first macrocell; (c) generating a carry-out signal from either the carry-in signal or a product term output, wherein selection of either the carry-in signal or the product term output is based upon the sum of products output corresponding to the first macrocell; and (d) continuing the carry cascade such that the carry-out signal from the first macrocell becomes a carry-in signal for the second macrocell, a carry-out signal for the second macrocell becomes the carry-in signal for the third macrocell, and so on, wherein the carry generation at each macrocell is based upon its corresponding sum of products output.
Use of the same reference symbols in different figures indicates similar or identical items.
The present invention recognizes that the pre-determined ratio of memory vs. logic blocks in conventional PLDs and FPGAs is often undesirable for particular user applications. It is always challenging to determine the ratio of memory vs. logic desired by users. The desired ratio varies, e.g., more than 50% or less then 10%, depending on particular system applications.
A programmable device with configurable generic logic blocks and methods of using the programmable device are provided in accordance with the present invention. In one embodiment, the programmable device comprises a microchip with a plurality of generic logic blocks (GLBs). The programmable device may be a general purpose, non-volatile, wide-input programmable device with super high gate densities, e.g., over one million system gates.
The structure of each generic logic block may be substantially similar, but each generic logic block is configurable to act as a complex programmable logic device (CPLD), an arithmetic unit, a configurable random access memory (RAM), a dual-port RAM, a configurable read only memory (ROM), a configurable content addressable memory (CAM), a configurable first-in-first-out memory (FIFO) and/or a configurable switch. The generic logic blocks improve data processing capabilities, flexibility, scalability and speed of a PLD and use a smaller footprint (less space on a microchip).
In one embodiment, a programmable device has one or more generic logic blocks that may be configured to perform a first function, while one or more other generic logic blocks are configured to perform a second function. For example, one generic logic block in the programmable logic device is configured as a RAM while two other blocks are configured as a FIFO, where all three blocks have the same structure. A user application may configure the generic logic blocks of the PLD for any ratio of logic, RAM, ROM, CAM, FIFO and switch functions. Thus, system functions can be implemented with any configurable ratio of memory vs. logic. The configurable generic logic blocks provide a homogeneous design environment that adapts to various system design challenges.
In accordance with the invention, each versatile generic logic block uses a configurable AND array to implement logic, SRAM, dual-port SRAM, ROM (RAM with the write port deactivated), FIFO, CAM or dynamic connectivity, cross point switch functions with minimum on-chip logic. In one embodiment, the AND array comprises low-power complementary metal oxide semiconductor (CMOS) circuits with n-channel MOS (NMOS) gates. In another embodiment, the AND array comprises one or more sense amplifiers.
In one embodiment, the CMOS circuits in the AND array advantageously consume or dissipate substantially zero direct current (DC) power. In contrast to PLD devices with sense amplifiers, the CMOS circuits in a programmable device according to the present invention reduce the memory cell size, increase memory cell margins and decrease memory cell stresses, allow the device to have a higher density of gates, reduce direct current and voltage requirements and provide speed advantages that scale with semiconductor processes.
In another embodiment, the programmable device is coupled to input/output cells that consume or dissipate substantially zero direct current (DC) power.
In one embodiment, each generic logic block in the programmable device may be configured to be a dual-port RAM with a first port and a second port. Each port is configured to write data to or read data from the memory.
In one embodiment, each generic logic block further comprises internal hardwired control logic, such as multiplexers, address decoders and/or polarity control circuits. In one embodiment, all control logic is implemented in hardware and are self-contained within each generic logic block. The generic logic block advantageously does not require additional logic to perform CPLD, dual-port RAM, ROM, FIFO, CAM and switch operations.
In one embodiment, the programmable device further comprises hardwired control logic coupled to one or more of the generic logic blocks. The hardwired control logic is configured to control one or more generic logic blocks.
In one embodiment, the programmable device may cascade product terms from one generic logic block to another generic logic block.
In one embodiment, each generic logic block in the programmable device comprises a plurality of configurable macrocells. Each macrocell may be configured to operate according to one of the above-mentioned functions: CPLD, arithmetic, RAM, ROM, CAM, switch or FIFO.
In one embodiment, the programmable device further comprises product term sharing arrays configured to share product terms between macrocells.
In one embodiment, the programmable device further comprises product term chain logic configured to expand product terms from two or more macrocells into a chain of product terms.
In one embodiment, each generic logic block may be configured to perform memory functions and input routing. Input signals from a global routing resource routed to a memory module (configured in the generic logic block) should be in a pre-determined sequence or order, i.e., received on specific data input pins and address input pins. In contrast, when a generic logic block is performing logic, input signals from a global routing resource may not be required to be in a predetermined sequence, and thus enhanced input routing may not be needed. The configurable input routing resources in accordance with the present invention meets the input routing requirements of a memory, avoids increasing routing overheads when the generic logic block performs logic, and reduces the die size of the programmable device because global routing resources may be reduced.
In one embodiment, each generic logic block is configurable to simultaneously perform two or more of the above-mentioned functions: CPLD, arithmetic, RAM, ROM, CAM, switch or FIFO.
In one embodiment, each generic logic block is configurable to function as a memory, such as a first-in-first-out memory, with programmable width modulation. Input and output port widths of the memory may be configurable, such that an input port has a width configured differently than the width of an output port. In addition, the width (size of each stored word) and depth (number of stored words) of the memory may be configurable.
In
The generic logic blocks 102A-102H may be referred to as ‘generic’ or ‘homogeneous’ because the structure of each generic logic block 102 is similar, but each block 102 may be separately configured to perform one or more functions, such as complex programmable logic, arithmetic, random access memory (RAM), read only memory (ROM), content addressable memory (CAM), a first-in-first-out memory (FIFO) or switching. Similarly, each macrocell 104 (
In one embodiment with a one-level general routing pool 110, the routing wires from each macrocell 104 and input pad 242 (
Two-Level Routing or Segment Routing
High Performance ‘Zero Power’ Product Term Logic
A sense amplifier requires DC current to sense a memory cell. The more current provided to the sense amplifier, the faster the sense amplifier performs. Thus, there is a tradeoff between performance and power consumption. For example, a sense amp may be designed with a 1-ns product term delay and consumes 800 uA of current. Another sense amp can be designed with 5-ns product term delay at 300 uA.
In one example of a memory circuit, there are 164×16=2,624 sense amplifiers. If a total ICC target is 1 Amp, an AC current target is 400 mA, a DC current target is 600 mA and substantially all DC current are from the sense amps, the circuit can only afford 285 uA per sense amp. This is based on only 80% of the sense amps being used in a customer application. The circuit may have a 5-ns propagation delay from the sense amps alone. This delay may be too long for some customer applications.
In order to make the circuit faster, the circuit may have a programmable option to turn on one or more sense amps at 1 ns, with a cost of 800 uA. Every sense amp programmed to such high performance takes current away from the ICC budget of 3 low current sense amps. A design software may have to calculate the total ICC DC current for each customer application.
The sense amp usage may vary in each customer application. Some applications use less PTs and some applications use more PTs. Customer applications that use less PTs can afford more high-performance, optioned sense amps. In a typical customer application, 50% of the PTs are used. This means all the PTs are slow, but only 10% of PTs are high speed for critical signals.
As described below, one embodiment of the GLB 102 in accordance with the present invention comprises CMOS circuits with zero DC current. Thus, the GLB 102 does not need to slow down performance to reduce current. Every product term is at 1-ns speed grade. A device with these GLBs 102 may simplify a customer's design process because the customer does not need to decide which signals are critical. Customers also do not need to accept slower PTs for most of their applications. The overall system performance is optimized. The advantages of a device with zero-DC-current CMOS circuits (described below) to generate product terms are more significant in a high density CPLD.
In one embodiment, the programmable devices 100 and 120 in
A ‘zero-power’ memory cell may have a configuration with transistors and a latch (or half-latch), along with certain signal timing and pre-charging methods, to prevent current from flowing through one or more transistors of the memory cell during normal operation and editing modes, such as programming or erasing.
In another embodiment, the programmable devices 100 and 120 of
Logic/Arithmetic Mode
The AND array 200 in
In
Each product term circuit 208 sums one or more inverted and/or non-inverted input signals from the ports 206A-206Z according to a particular product term (PT). Each product term circuit 208 outputs a PT to a demultiplexer 210, which outputs the PT to a 6-input OR gate 212, a 5-input OR gate 214 and/or a macrocell 104.
Each macrocell 104 in
RCSF Mode
Dual-Port RAM Mode
Macrocell
In one configuration, the surrounding logic in
In addition to the input lines shown in FIG. 2A and described above, the macrocell 104 of
The D register/latch 324 can be programmed to be a register or a latch. Outputs from the macrocell 104 are either registered (outputs from the D register 324) or combinatorial (outputs from the PTSA bypass line 220 or bypass line 338). The macrocell 104 of
The macrocell 104 of
Carry Generation and Carry Propagation
Any logic function may be represented by Boolean equations. A Boolean equation may be expressed as several product terms (PTs) OR'ed together. Many frequently-used functions are arithmetic in nature. The most significant bit (MSB) of an arithmetic output depends on the evaluation of the lower order bits. Implementing a MSB function in a conventional CPLD requires either wide-OR functions, or cascaded logic into multi-level logic units. A wide-OR function requires logic resources. Also, a conventional CPLD is not efficient with wide-OR functions because the CPLD has a limited total number of available product terms. Cascaded logic degrades performance (long process time because of multiple levels of logic).
According to the present invention, a flexible CPLD macrocell 104 (
In
The Carry In line/signal 302 and XOR gate 306, coupled with OR gate 214 and OR gate 212, of the macrocell 104 in
Product Term Sharing
The expanded PTSA 202 in
For example, a macrocell 104D (not shown) may use its own five PTs, share 15 PTs from three macrocells 104A-104C (
Each macrocell 104 does not have to share PTs from three macrocells 104 of each side. For example, each macrocell 104 may share PTs in patterns of every other macrocell 104 or a random pattern to maximize programmability.
Product Term Chaining
The 6-input OR gate 212 in
The sixth input line 335 of the 6-input OR gate 212 does not have to be connected to an N-7 macrocell. The sixth input line 335 of the 6-input OR gate 212 may be connected to any FT that is not accessible to a macrocell 104 through the PTSA 202. For example, if the product term sharing array 202 is arranged to share product terms among every other macrocell 104, each macrocell 104 may be chained to one or more N-1 macrocells.
Cascading Product Terms
The device 100 in
The first GLB 102A (
The four control PTs (CLOCK, CLOCK ENABLE, RESET, GPTOE) in
Split AND Array w/Split Product Terms
In one embodiment of RCSF mode, each of the first 36 PTs (0 to 35) formed by AND array circuits 402A-402Z in
Splitting the AND array outputs to form 72 split outputs 404A-404Z in
The split AND array 400 in
Carry Ripple/Arithmetic Support
Each GLB 102 in
For example, the macrocell 104 in FIG. 3 and surrounding logic may implement a 1-bit adder, such as A+B+CARRYIN=SUM+CARRYOUT. The five product term circuits 208A-208E, the five DEMUXes 210A-210E, the MUXes 300B, 334A, 300F, 334C, 300I, the 5-input OR gate 214, the 6-input OR gate 212, the expanded PTSA 202, the XOR gate 306, the XOR gate 312, MUX 334A and the paths between these components may be active to implement the adder function. The second product term circuit 208B receives inputs A and B and outputs a product term of A AND B. Any of the remaining product term circuits 208A, 208C-208E may implement two product terms of ˜AB and A(˜B). The 5-input OR gate 214 may output P=A XOR B=˜AB OR A(˜B). The XOR gate 306 outputs a SUM (SUM=A XOR B XOR CIN) and may transfer the SUM to an input of the 6-input OR gate 212 via MUX 334A. The output of the 6-input OR gate 212 may be routed to the macrocell flip flop 324 or the PTSA 202.
The PTSA 202 may route SUM to flip flops 324 in different macrocells 102 (
A first CARRYIN signal for an adder/subtractor may be 0, 1 or an external signal. If a PT drives a CARRYIN input signal, then the CARRYIN input signal is routed through an extra macrocell 102 (FIG. 2A). The extra macrocell 102 will always ‘generate’ a carry equal to the CARRYIN input PT, e.g., with the output of the 5-input OR gate 214 (
A GLB 102 (
If a macrocell 104 (
For a device 120 (
RAM Mode
A GLB 102 in
In one embodiment, the sets of input registers 502 and 504 comprise special registers that are not available for general logic use, while the set of output registers 506 comprises general macrocell registers that are available for general logic use. The first set of input registers 502 receives addresses and outputs the addresses to the RAM array 500 for data to be read according to a read address and read clock signals. The second set of input registers 504 receives data from an external source and outputs data to be written in the RAM array 500 according to a write address, data, write enable, and write clock signals. The set of output registers 506 receives data from the RAM array 500 and outputs the data according to a read clock signal. The first set of input registers 502 and the set of output registers 506 may used to synchronously read data from the RAM array 500. The GLB 102 in
In
Split AND Array for Memory and Input Routing
When the GLB 102 described herein is configured to perform memory functions, the GLB 102 uses a part of the AND array 200 in
The portion of the split AND array configured as a switch matrix 624A in
In
The split AND array switch matrix 624A has a read address (RA) output port 628A, a read clock (RCLK) output port 630, a read clock enable (RCLKEN) output port 632, a reset output port 634, a write clock enable (WCLKEN) output port 636, a write clock (WCLK) output port 638, a write address (WA) output port 640A, a write data (WD) output port 642A, a write enable (WE) output port 644 and a write chip select (WCS) output port 646. In one embodiment, product terms 160-163 are not split product terms and are used to implement regular AND product terms.
In another embodiment, the GLB 102 does not have a split AND array switch matrix 624A. Instead, an enhanced GRP 110 delivers input signals in proper sequence and order to the other components in FIG. 6A. In this embodiment, the memory 600A may he larger and comprise all of the product term circuits 208A-208Z in FIG. 2A.
The optional MUXes and logic 648 may provide a selection between inverted control signals and non-inverted control signals. The optional MUXes and logic 648 may provide a selection between clock signals generated by the split AND array switch matrix 624A and a global clock signal GCLK on line 686. The optional logic 648 may comprise AND gates and OR gates, for example, to perform a logical AND operation between signals WE and WCS or perform a logical OR operation between signals RESET and global reset.
The D flip flops/registers 674 may buffer some of the output signals of the split AND array switch matrix 624A. The write address decoder logic 662A passes WA, WE, WCS signals to the SRAM 600A. The read decoder logic 664A passes RA signals to the SRAM 600A. The SRAM 600A receives write data via an input line 610A from the split AND array switch matrix 624A and outputs read data via an output line 622A to a destination, such as macrocells, or an external source, such as a local routing pool or the GRP 110 (FIG. 1A).
The split AND array switch matrix 624B comprises the split product term circuits 408A-408Z, 410A-410Z and product term circuits 402A-402Z discussed above with reference to FIG. 4B and may comprise additional logic. The split AND array switch matrix 624B receives signals from the GRP 110 via a 68-bit input line 626. The split AND array switch matrix 624B has a read address (RA) output port 628B, a read clock (RCLK) output port 630, a read clock enable (RCLKEN) output port 632, a reset output port 634, a write clock enable (WCLKEN) output port 636, a write clock (WCLK) output port 638, a write address (WA) output port 640B, a write data (WD) output port 642B, a write enable (WE) output port 644 and a write chip select (WCS) output port 646. Product terms 160-163 are not split product terms and are used to implement regular AND product terms.
The optional MUXes and logic 648 may provide a selection between inverted control signals and non-inverted control signals. The optional MUXes and logic 648 may provide a selection between clock signals generated by the split AND array switch matrix 624B and a global clock signal GCLK on line 686. The D flip flops/registers 674 may buffer some of the output signals of the split AND array switch matrix 624B. The write address decoder logic 662B passes WA, WE and WCS signals to the SRAM 600B. The read decoder logic 664B passes RA signals to the SRAM 600B. The SRAM 600B receives write data via an input line 610B from the split AND array switch matrix 624B and outputs read data via an output line 622B to a destination, such as macrocells, or an external source, such as a local routing pool or the GRP 110 (FIG. 1A).
The six memory configurations in
Dual-Port RAM Mode
The split AND array switch matrix 702A in
The optional MUXes and logic 744 may provide a selection between inverted control signals and non-inverted control signals. The optional MUXes and logic 744 may provide a selection between clock signals generated by the split AND array switch matrix 702A and a global clock signal GCLK on line 686. The optional logic 744 may comprise AND gates and OR gates, for example, to perform a logical AND operation between signals WE and WCS or perform a logical OR operation between signals RESET and global reset.
The D flip flops/registers 776 may buffer for some of the output signals of the split AND array switch matrix 702A. The D flip flops/registers 776 may be reset by a global reset signal on line 792. The write address decoder logic 704A passes control and address signals to the SRAM 700A. The read decoder logic 706A passes control and address to the SRAM 700A.
The SRAM 700A receives write data via two input lines 736A, 738A, from the split AND array switch matrix 702A and outputs read data via two output lines 740A, 742A to a destination, such as macrocells, or an external source, such as macrocell registers, a local routing pool or the GRP 110 (FIG. 1A).
The split AND array switch matrix 702B in
The optional MUXes and logic 744 may provide a selection between inverted control signals and non-inverted control signals. The optional MUXes and logic 774 may provide a selection between clock signals generated by the split AND array switch matrix 702A and a global clock signal GCLK on line 686. The D flip flops/registers 776 may buffer some of the output signals of the split AND array 702A. The D flip flops 776, 778, 780, 786, 788, 790 may be reset by a global reset signal on line 792. The write address decoder logic 704B passes control and address signals to the SRAM 700B. The read decoder logic 706B passes control and address signals to the SRAM 700B.
The SRAM 700B receives write data via two input lines 736B, 738B, from the split AND array switch matrix 702B and outputs read data via two output lines 740B, 742B to a destination, such as macrocells, or an external source, such as macrocell registers, a local routing pool or the GRP 110 (FIG. 1A).
Data and Address Ports with Configurable Widths
The widths of the data and address ports of the RAM arrays 600A, 600B, 700A, 700B in
As another example, the RAM array 700A in
If the data bus width of a read port and a write port (in either a single port RAM 600A (
MSB=(current_address+1)×port_width−1
LSB=current_address×port_width
Using the example above, if port A (total addresses: 0 to 511) writes a 32-bit word at an arbitrary address location 88, then the MSB of that word goes to the physical RAM bit location (88+1)×32=2847, while the LSB of the word goes to location 88×32=2816. The bits between the MSB and LSB go between locations 2816 and 2847. Port B (total addresses: 0 to 2047) would read four consecutive 8-bit words from locations 88×4+0=352, 88×4+1=353, 88×4++2=353, 88×4+3=354, 88×4+4=355, and receive the same data as written by port A. The least significant 8 bits of port A map to the 8-bit word at location 88×4=352, while the most significant 8-bits map to the 8-bit word at 88×4+3=355. The logical to physical conversion equation above shows that these four words fall on appropriate physical RAM bit locations. For example, port B's location 352's MSB is at physical RAM bit location (352+1)×8=2824, while LSB is at location 2816. These locations represent the least significant 8-bits of port A's 32-bit word at location 88. For a dual-port RAM mode, the total SRAM capacity is cut in half, and there are half as many words for each data bus width.
The second set of ports comprises a write enable (WEB) input port 715, a write clock (WCLKAB) input port 717, a read/write address (RWB) input port 719, a write chip select (WCSB) input pot 721, a write data (WDB) input port 723, an enable input port (ENB) 725 and a reset B input port 727. As shown in FIG. 7C, the ‘dual-port’ RAM 700 has a first set of independent read and write ports and a second set of independent read and write ports.
ROM Mode
The GLB 102 described herein may be configured as a read only memory (ROM) by implementing a lock-in bit to prevent write access. The GLB 102 acting as a ROM may be loaded with initial configuration data upon start-up, which cannot be altered later.
CAM Mode
As described above, a GLB 102 of
The control logic 806 in
The portion of the split AND array configured as a switch matrix 810 in
The optional MUXes and logic 836 may provide a selection between inverted control signals and non-inverted control signals. The optional MUXes and logic 836 may provide a selection between signals generated by the split AND array switch matrix 810 and a global clock signal GCLK on line 686. The optional logic 836 may comprise AND gates and OR gates, for example, to perform a logical AND operation between signals WE and WCS or perform a logical OR operation between signals RESET and global reset.
The D flip flops/registers 862 buffer some of the output signals of the split AND array 810. The D flip flops 862 may be reset by a global reset signal on line 792. The read/write address decoder logic 812 passes control signals, such as clock signals, and address bits for data to be read or written to the CAM array 800.
The compare data encoder in the logic 812 encodes registered CAM compare data into 3 states (0, 1 or DON'T CARE) depending on the mask register bits from the register 864. The write data encoder in the logic 812 encodes registered CAM update data into 3 states (0, 1 or DON'T CARE) depending on the mask register bits from the register 864 and the WR_DON'T_CARE signal.
The CAM array 800 receives compare data and write data from the compare data encoder and the write data encoder of the logic 812. The data stored in the CAM array 800 may include data and any user-defined validity bits or data classes.
The CAM array 800 outputs up to 128 bits of unencoded data to the match flag logic 874 and the CAM output encoder 876. The match flag logic 874 determines the number of matching entries in the CAM array 800 and outputs a MATCH_OUT signal and a MUL_MATCH_OUT signal to a destination, such as macrocells, or an external source, such as macrocell registers, a local routing pool or the GRP 110 (FIG. 1A). The CAM output encoder 876 priority encodes addresses of any entries in the CAM array 800 that match data from the compare data encoder of the logic 812. The CAM output encoder 876 outputs the 7-bit address of the entry with the highest priority (entry 0 is the lowest address and has the highest priority).
In one configuration, a reset signal resets the output registers but does not affect the CAM array 800. The clock enable (CE) signal disables the clock inputs to all registers 862-872. The write enable (WEN) signal in conjunction with any depth cascading chip selects (CS) enables writing into the CAM array 800.
FIFO Mode
The split AND array switch matrix 900 in
A first set of the MUXes 922-928 provide a selection between inverted control signals and non-inverted control signals. A second set of MUXes 929, 930, 936, 940 provide a selection between signals generated by the split AND array switch matrix 900 and a global clock signal GCLK on line 686. A third set of MUXes 950, 952 provide flag values from macrocells 104 (FIG. 2B). The D flip flop 946 provides a register or buffer for write data from the split AND array switch matrix 900.
The first and second flag logic circuits 944, 948 may be implemented as separate units or as a single unit. The first and second flag logic circuits 944, 948 may be reset by a global reset signal on line 792 or a block-level reset on line 914. The first flag logic circuit 944 outputs two programmable flags, ALMOST EMPTY (Y) and EMPTY, to external logic, such as the macrocells 104 (FIG. 2B). In one embodiment, the first flag logic circuit 944 (
The second flag logic circuit 948 outputs programmable FULL (also called TC for terminal count) and ALMOST FULL (X) flags to external logic, such as macrocells 104 (FIG. 2B). In one embodiment, the second flag logic circuit 948 (
The FULL flag is programmable from 1 to a maximum number locations in the FIFO 900. The ALMOST FULL and ALMOST EMPTY flags are programmable from 1 to FULL—1. If a user tries to write to the FIFO 904 after a FULL flag is active, or tries to read from the FIFO 904 after an EMPTY flag is active, then the internal pointers of the FIFO 904 are frozen at their last active value, and no read or write operation is performed.
The read and write address decoder logic 902, 906 passes control signals, such as clock signals, and address bits for data to be read from or written to the FIFO 904. The FIFO 904 receives data to be written from the second flag logic circuit 948. The FIFO 904 outputs up to 32 bits of data to an external source, such as macrocell registers, a local routing pool or the GRP 110 (FIG. 1A).
The FIFO 904 may have differently-configured data bus widths on the FIFO's read and write ports. For example, in one embodiment, the FIFO 904 has a 32-bit write port and an 8-bit read port. In this embodiment, the FULL and ALMOST FULL flags refer to 32-bit words being written, and the EMPTY and ALMOST EMPTY flags refer to 8-bit words being read. For example, assuming the EMPTY and ALMOST EMPTY flags are programmed to be 2, a first 32-bit write will clear the EMPTY and ALMOST EMPTY flags because the FIFO 904 has 4 words to be read. But the ALMOST FULL flag will not be asserted because there is only one written word in the FIFO 904. Assuming a second 32-bit write sets the ALMOST FULL flag, the FIFO 904 now has 8 words to be read. Reading 6 words would set the ALMOST EMPTY flag, and reading 8 words would set the EMPTY flag.
The order of the 8-bit words read from the FIFO 904 may be changed by software, hardware or both. For example, either software or hardware may change the order of data received by input data pins of the FIFO 904 or the order of data sent by output data pins of the FIFO 904.
When the first and second flag logic circuits 944, 948 (FIG. 9A), respectively, receive a reset signal (either a block-level reset via line 914 or a global reset via line 792), the read and write pointers (RP and WP) in the first and second flag logic circuits 944, 948 are reset to “0,” the first location in the FIFO 904. In one embodiment, the read and write pointers may vary in width from 9 to 14 bits. At a first valid RCLK/WCLK clock edge from output ports 910, 920 (FIG. 9A), the read and/or write pointers are advanced to location “1” to point to the next location in the FIFO 904.
The subtractor P1 in
When a reset signal is sent to the macrocell registers 970-976 that store the flags, the FULL and ALMOST FULL flags are de-asserted (set to 0), and the EMPTY and ALMOST EMPTY flags are asserted (set to 1). In other words, the inverse of the EMPTY and ALMOST EMPTY flags (!EMPTY and !ALMOST_EMPTY are set to 0. The third and fourth comparators 966, 968 have inverters at their output ports.
The first flag logic circuit 944 in
Some of the macrocells 104 (
Programmable Width Modulation
As described above, the GLB 102 may be configured as an SRAM, a dual-port SRAM, a CAM, a switch or a FIFO with programmable width modulation. For example, a traditional FIFO has a read port and a write port with essentially the same data width. ‘FIFO depth’ is defined as a number of words that the FIFO can store before a FULL flag is asserted. An EMPTY flag is asserted if there are more words read then written in the FIFO.
A FIFO with programmable width modulation in accordance with one aspect of the invention has a read port data width that is independent of a write port data width. For example, if the write port is narrower than the read port, the FIFO control logic packs written data before the data is read. If the write port is wider than the read port, the FIFO control logic unpacks written data before the data is read. The FIFO may also compare read word count and write word count to determine FULL and EMPTY flag statuses.
Switch Mode
The split AND array switch matrix 1000 in
A first set of the MUXes 1016-1022 provide a selection between inverted control signals and non-inverted control signals. A second set of MUXes 1023-1026 provide a selection between signals generated by the split AND array switch matrix 1000 and a global clock signal GCLK on line 686. The D flip flops 1027-1031 provide registers or buffers for signals from the split AND array switch matrix 1000 to the shadow memory 1004 and the switch 1006. The data inputs and outputs may be registered or unregistered.
The write address decoder logic 1002 passes WE, A_IN and A_OUT to the shadow memory 1004 and the main memory 1005. The A_IN signal selects one or more input bits of DATAIN for the switch 1006 to route. The A_OUT signal specifies a routing configuration that determines an output routed by the switch 1006. The shadow memory 1004 and the main memory 1005 store routing configurations (A_OUT). The DATAIN register 1027 is controlled by the DCLK and DCE signals. The address register 1030 for A_IN and A_OUT is controlled by the ACLK and ACE signals. In one embodiment, the DCLK, DCE, WE, UPDATE and ACE inputs are not used. The WE, UPDATE and ACE input ports may be tied to 1 if they are unused.
The switch block 1006 outputs up to 32 bits of data, according to a routing configuration, to an external source, such as macrocell registers, a local routing pool or the GRP 110 (FIG. 1A).
When the switch block 1006A receives an asserted FLOWTHROUGH signal, each 1-bit input pin outputs its value (either 0 or 1) to a corresponding 1-bit output pin. For example, input pin 0 outputs its value to output pin 0, input pin 1 outputs its value to output pin 1, etc.
In a MULTICYCLE MULTICAST mode, the switch blocks 1006A-1006C (
In one embodiment, an update of WE, A_IN and A_OUT from the shadow memory 1004 to the main memory 1005 uses one clock cycle, and the switch 1006 may be configured while the main memory 1005 is operating the connections.
The LRP 1054 in
In FLOWTHROUGH mode, the memories 1004, 1005 do not have to be updated. Thus, the switch 1006 conforms to the FLOWTHROUGH mode as long as the FLOWTHROUGH signal is asserted. When the FLOWTHROUGH signal is de-asserted, the switch 1006 reverts back to the routing configuration stored in the main memory 1005.
Product Term Circuit in the AND Array
In other embodiments, the product term circuit 208 may comprise less or more gates than those shown in FIG. 11A. For example, although 34 NAND gates 1100A-1100Z are shown in a first stage of the product term circuit 208 in
In one embodiment, each NAND gate 1100 in the first stage of the product term circuit 208 in
In one embodiment, out of 164 product term circuits 208, 36 product term circuits 208 (configured for input routing or product term logic) comprise CMOS circuits, such as the CMOS circuit 1440 in
In one embodiment, the second, third, fourth and fifth stages of NOR gates 1102A-1102Q, 1106A-1106C, and NAND gates 1104A-1104F, 1108 are standard NOR and NAND gates.
In another embodiment, each product term circuit 208 in
A CMOS Circuit
The row write word line 1144 and the row write bit line 1146 in
A product term circuit 208 in
In
In
Split AND Array and Control Logic Overview
As shown in
In one embodiment, all structures shown in
In one embodiment, the RCSF AND array portion 1200 in
In one embodiment, the input routing AND array portion 1204 comprises 32 AND array product term circuits 208, which are described herein with reference to
The WWL/RROW drivers 1202 in
The column drivers 1206 in
The RCSF clock MUXes and RCSF input registers 1222 in
The CAM control logic 1208 in
The data out multiplexer (DOUTMUX) 1210 in
The architecture configuration memory cells (ARCH) 1212 in
The FIFO control logic 1216 in
CPLD/RAM Circuit with Control Logic
The additional logic comprises row drivers 1304A, 1304B, a read x-address decoder (‘x-decoder’) 1310, mode select circuits 1312A, 1312B, a write address decoder and write word line (WWL) driver 1314, a bit line driver or drivers 1316, select circuits 1350, 1352, a read column-address decoder (‘y-decoder’) 1318, an array 1320 of OR gates, another mode select circuit 1322, a PTSA 202 and a set of macrocells 104A-104Z (FIG. 2A). Each row driver 1304 may comprise a plurality of buffers and inverters. In one configuration, for example, the row driver 1304A comprises 64 buffers and 64 inverters, and the row driver 1304B comprises 4 buffers and 4 inverters.
The mode select circuit 1312A receives configuration bits for CPLD mode or RAM mode from the circuit 1312B and then selects either the output of the row driver 1304A (CPLD mode) or the output of the read address decoder 1310 (RAM mode) to send to the AND array portion 1302A via lines/bus 1324A.
In one embodiment, to perform CPLD functions, the row drivers 1304A and 1304B receive and pass 68 input bits and their 68 inverted input bits (136 total) to the mode select circuits 1312A and 1312B. The mode select circuits 1312A and 1312B pass the 136-bit output of the row drivers 1304A and 1304B to the AND array portions 1302A and 1302B via lines/bus 1324A and 1324B to implement product term logic. The combined AND array portions 1302A and 1302B send 128 outputs to the array 1320 of OR gates, and the mode select circuit 1322 passes the outputs of the array 1320 to the PTSA 202 and 32 macrocells 104A-104Z (FIG. 2A). The array 1320 of OR gates may represent an array of 32 6-input OR gates 212 (
A second AND array portion 1332 (e.g., an input routing pool (IRP)) takes the same 68 inputs from the GRP and generates another 32 product terms for a total of 160 product terms to send to the array 1320 of OR gates.
In RAM mode, one configuration of the AND array portion 1302A in
To perform a RAM read function, the read address decoder 1310 receives and decodes a 7-bit address to select one of 128 possible memory row entries (27=128) in the AND array portion 1302A to be read. The read address decoder 1310 sends one active signal (‘1’) on one of 128 lines to the mode select circuit 1312A and ‘0s’ on the other 127 lines to the mode select circuit 1312A. Thus, the read address decoder 1310 may be described as a ‘1-hot’ decoder circuit. In another configuration, the read address decoder 1310 is configured to send one ‘0’ on one of 128 lines to the mode select circuit 1312A and send ‘1s’ on the other 127 lines to the mode select circuit 1312A.
The mode select circuit 1312A sends the output of the read address decoder 1310 to the AND array portion 1302A via lines/bus 1324A. The row input line 1142 in
The y-decoder 1318 allows the output of the AND array portion 1302A to be configurable to various data widths. In one configuration, the y-decoder 1318 is configurable to act as a 4-to-1 multiplexer (MUX) to output 32 of 128 bits at a time. In other configurations, the y-decoder 1318 is configurable to act as an 8-to-1 MUX to output 16 bits at a time, a 16-to-1 MUX to output 8 bits at a time, a 32-to-1 MUX to output 4 bits at a time, a 64-to-1 MUX to output 2 bits at a time, or an 128-to-1 MUX to output 1 bit at a time.
The mode select circuit 1322 selects either the outputs of the y-decoder 1318 (RAM mode) or the outputs of the OR gate array 1320 (CPLD mode) to send to the PTSA 202 and macrocells 104A-104Z.
Another AND array 1332 (FIG. 13), also called an input routing pool (IRP), generates 64 outputs from 68 GRP inputs. The 64 outputs are used by the circuit 1300 configured as a RAM as address, input data, control signals and other things.
To perform a RAM write function, the write address/write word line (WWL) driver 1314 receives and decodes a 7-bit address to select one of 128 possible memory row entries (27=128) in the AND array portion 1302A to write data. The write address decoder 1314 sends a single (‘1’ or ‘0’) on one of 128 lines 1326 to the AND array portion 1302A via the select circuit 1350, which receives a WWL signal from an address shift register (ASR). The row write word line (RWWL) 1144 in
The bit line driver 1316 receives a 32-bit word and a 2-bit address (00, 01, 10, 11) to select which group of 32 bits to write at one time in a row of 128 SRAM cells (see
A Dual-Write Port SRAM Cell
In one embodiment, the product term circuits 208 (
In one embodiment, SRAM cells such as cell 1400 in
Data may be written on a row-by-row basis by selecting one of the write word lines AWWL 1424A or BWWL 1424B and applying data to the corresponding sets of bit lines ABL 1422A, /ABL 1422C or BBL 1422B, /BBL 1422D. The column select line 1418 is driven by a column address decoder (Y decoder) 1612B (
In one embodiment, dual port read is achieved by reading from each port's write word lines and bit lines. This embodiment may have sense amplifiers attached to bit line A and bit line B.
In another embodiment, the read path NAND 1414 and row driver line for each port is duplicated. Since the SRAM cell 1400 in
In other embodiments, the product term circuit 208 in
A Dual Write Port Cell with Shadow Memory
In one embodiment, the circuit 1500 may further comprise two NAND gates 1520 and 1522.
Dual-Port RAM Circuit
The operation of the dual-port RAM circuit 1600 in
In one embodiment, the two AND array portions 1602A, 1602B are not separated physically but are shown as separate portions for the purposes of illustration. The two AND array portions 1602A, 1602B may comprise even and odd rows of CMOS circuits, respectively, such as the CMOS circuits 1400, 1500 shown in
In RAM-mode operation, the port A x-decoder 1608A may decode port A addresses and write the same data to both AND array portions 1602A, 1602B simultaneously. The port B x-decoder 1608B may decode port B addresses and write data to both AND array portions 1602A, 1602B simultaneously. Likewise, the port A bit line driver 1610A may activate ‘A’ bit lines in both AND array portions 1602A, 1602B simultaneously. The port B bit line driver 1610B may activate ‘B’ bit lines in both AND array portions 1602A, 1602B simultaneously. Thus, the AND array portions 1602A, 1602B will both store the same data.
If the port A and port B x-decoders 1608A, 1608B and bit line drivers 1610A, 1610B attempt to write to the same memory location in one of the AND array portions 1602A, 1602B, then the data written is not valid. The circuit 1600 may assert an error flag.
In another embodiment, the port A x-decoder 1608A writes data to the AND array portion 1602A, and the port B x-decoder 1608B writes data to the AND array portion 1602B. If data is written to the AND array portion 1602A, then the data is copied to the AND array portion 1602B. Similarly, if data is written to the AND array portion 1602B, then the data is copied to the AND array portion 1602A.
Two independent port y-decoders 1612A, 1612B in
In CPLD mode, the mode select 1606B may pass 64 buffered inputs to the AND array portion 1602B. The mode select 1606A may pass 64 inverted inputs to the AND array portion 1602A. The two AND array portions 1602A, 1602B produce outputs to an array 1614 of AND gates, where each AND gate may be the AND gate 1108 in
FIFO Circuit
In one embodiment, the read counter 1706 and half of the FIFO flag logic 1710 in
The read counter/pointer 1706 in
The SRAM portion 1302A in
A RCLK line, WCLK line, and write data line may all connected to IRP 1332. The mode select 1322 in
The FIFO flag logic 1710 compares the values of the read counter 1706 and write counter 1708. For example, the FIFO flag logic 1710 may subtract the read counter 1706 from the write counter 1708 to determine an actual depth of the memory 1302A. The FIFO flag logic 1710 may then compare the actual depth of the memory 1302A with a pre-determined depth value to determine whether the memory 1302A is full. The FIFO flag logic 1710 in
The FIFO flag logic 1710 may have timing circuits to handle two time domains, one time domain for the read counter 1706 and another time domain for the writer counter 1708, and avoid oscillations of the flags. Thus, the FIFO flag logic 1710 may protect the integrity of data stored in the memory 1302A, i.e. prevent overwriting or losing data. The FIFO flag logic 1710 may output flag statuses to other circuits.
CAM Circuit
Each CAM ‘bit’ is represented by two SRAM bits in the CAM AND array portion 1302A. When writing a word A, the CAM circuit 1800 writes each bit AI to its true and complement forms AI and /AI (also called “AIbar”) to two SRAM cells in the CAM AND array portion 1302A. As shown in
The CAM compare function may be implemented as a CPLD product term (one column or one product term circuit 208
The CAM write operation may be implemented similar to a dual-port SRAM described above, except for the differences described herein. Assuming CAM words are compared PT-by-PT, words may be written into the CAM AND array portion 1302A in the column direction (column-by-column) because PTs are aligned in the column direction. One way to update/write a column is to update row-by-row and enable one column select line 1418 in FIG. 14. This process will take 128 cycles for one CAM word update.
To speed up the CAM update process, a CAM circuit may select and write to all rows in a enabled column that should be written with a ‘1’ in the CAM AND array portion 1302A in one cycle. Then the CAM circuit selects and writes to all the rows in a enabled column that should be written with a ‘0’ in the CAM AND array portion 1302A in the next cycle. This reduces a CAM write operation to two cycles.
In one embodiment, the CAM write operation may be done in one cycle, and a multi-match may be done in one cycle. To update all rows in an enabled column in one cycle, the row drivers 1304A, 1830 and the x-address decoder 1808 in
Cross-Point Switch
A cross-point switch (
Mask Programming
To reduce manufacturing costs and improve performance, a manufacturer may produce a device with mask-programmed circuits derived from a field-programmable device 100 (described above, e.g.,
The designer prototypes the field-programmable device 100 (with multiple GLBs 102) to implement a particular function or set of functions, such as functions in a cell phone. The designer may determine that the particular function needs, for example, X % of the GLBs 102 in the device 100 to perform PT logic, Y % of the GLBs 102 to perform RAM, and Z % of the GLBs 102 to perform FIFO. The designer informs the manufacturer that the designer desires a device with a desired ratio of PT logic vs. memory functions (e.g., X % of the GLBs 102 in the device 100 to perform PT logic, Y % of the GLBs 102 to perform RAM, and Z % of the GLBs 102 to perform FIFO).
The designer may provide additional instructions to the manufacturer on how a device desired by the designer should be programmed and configured. For example, the designer may provide a specific configuration fusemap that specifies each pin's function and routing information inside the device 100. This fusemap is used in the field to program the device 100. The designer gives this fusemap to the manufacturer and asks the manufacturer to make a mask that follows the information in the fusemap.
Based on the designer's instructions, the manufacturer creates a second device with one or more “mask-programmed” sections that are pre-programmed in the manufacturer's factory, and are not programmable in the field. Specifically, the manufacturer replaces one or more field-programmable circuits in the device 100 with mask-programmed circuits to perform the function(s) specified by the designer. The manufacturer uses mask steps (dedicated to a function specified by the designer) in a multi-step IC device manufacturing process to create the mask-programmed circuits. Since mask-programmed circuits cannot be changed in the field, the mask-programmed part of the second device is essentially customized for a given designer. The relationship between the first device 100 and the second device may be similar to the relationship between an EEPROM and a Mask ROM.
The combination of the first programmable device and the second device with a mask-programmed section offer a greater value for the designer than either device by itself. Because the second device includes a mask-programmed section, the second device is less expensive to manufacture than the first device. The mask programmed circuits may be much smaller and faster than field-programmable circuits. Thus, mask-programmed circuits offer a cheaper and higher performance part to a designer.
In one embodiment, all of the programmable circuits in the first device 100 are converted to pre-programmed, non-field-programmable, mask-programmed circuits to form the second device. For this embodiment, the manufacturer is able to replace all switches that controlled by fuse with metal connection/disconnection. The manufacturer may be able to eliminate all non-volatile cells.
In another embodiment, the finished second device may comprise a combination (or ratio) of one or more pre-programmed, non-field-programmable, mask-programmed circuits and one or more field-programmable circuits, such as the configurable product term circuits of
In both embodiments above, the manufacturer can generate a smaller die with reduced circuits. The new die should have higher performance because of the smaller die and several configuration switches are replaced by metal connections. Both embodiments will work for designers who use PLD during prototyping and change to ASIC during production.
ASIC I/O Module
The device 100 described above may further comprise one or more programmable input/output (I/O) modules configured to allow the device 100 to select an input/output interface standard (e.g., output voltage) from a plurality of predetermined interface standards, such as 5.0-volt and 3.3-volt output voltages, differential output voltages (the output signal is the difference in voltage between two pins), open drain and other output voltage standards known to those of ordinary skill in the art.
In one embodiment, the programmable input/output (I/O) module comprises a plurality of advanced I/O cells (ASIC I/Os), where each advanced I/O cell may be configured to use one I/O interface standard while another advanced I/O cell is configured to use another I/O interface standard. Thus, various I/O interface standards are used simultaneously at different I/O cells. In another embodiment, the programmable input/output (I/O) module comprises one or more low voltage differential signal (LVDS) input/output modules.
Other Components: Phase Locked Loop, State Machine
The programmable device 100 may further comprise an on-chip phase locked loop (PLL) to enhance clock functionality, a state machine to allow an external CPU to program the device, a state machine to allow JTAG programmability and/or a state machine to allow In-System Programmability (ISP).
Package and Pinout Migration
According to one aspect of the invention, a manufacturer may provide a family of integrated circuit devices, where each integrated circuit device comprises a different number of GLBs, i.e., a different number of macrocells, where one GLB corresponds to 32 macrocells in one embodiment. For example, the manufacturer may offer devices with 256, 368, 384, 512, 768, 1024, 1536 or 2048 macrocells. The number of macrocells in a device may be referred to as the “density” or “capacity” of the device.
An integrated circuit device may be assembled into a “package.” A device with a particular density (e.g., 512 macrocells) may be assembled in a number of different package configurations. Each package configuration may have a different number of total pins, such as 160, 208, 272, 388 or 484 pins. Some of the pins are I/O pins and some of the pins are control signal pins (e.g., VCC, VSS, CLK, control). For example, a device with a particular density (e.g., 512 macrocells) may be assembled in packages with 160, 208, 272, 388 or 484 pins. For each of these packages, a various number of pins are assigned to function as 10 pins, such as 104, 156, 196, 212, 256 or 304 I/O pins.
In addition, each package configuration with a set number of pins may be configured to house devices with different densities, one at a time. This may be called “package migration.” To house devices with different densities (one at a time) in the same package configuration, all devices preferably have the same pinout definition or specification to accommodate specific signals (e.g., VCC, VSS, CLK, CONTROL, global signals, and macrocell I/Os) assigned to pins at specific locations of the package. For example, if pin 38 on a 272-pin package is assigned to VCC, then all devices that may be housed in that package are configured to accommodate VCC at pin 38. If pin N on a package is assigned to I/O, then pin N is I/O or “not connected” for a device assembled in the package. In one embodiment, all signals such as VCC, VSS, CLK, CONTROL, etc. are at identical pins on the devices. “Pinout migration” refers to devices with various densities and an identical pinout, which can be housed in one package configuration.
For example, a package with 388 pins can house a device with 512 macrocells, where 256 of the 388 pins are used as I/O pins. Alternatively, the same 388-pin package can house a device with 768 macrocells, where 304 of the 388 pins are used as I/O pins. The packaged device with 512 macrocells has more “not connected” pins than the packaged device with 768 macrocells because the device with 768 macrocells uses more pins on the package as I/O pins. Thus, a user can use the 388-pin package as a socket and swap or replace a device with 768 macrocells with a device with 512 macrocells and use the extra “not connected” pins for other inputs/outputs.
These packages allow a user to migrate (i.e., “drop-in,” replace or “swap”) a higher or lower capacity integrated circuit device without redesigning the layout of a circuit board or any support chip. For example, a first device with 512 macrocells may be housed in either a 484-pin package or a 388-pin package. A second device with 256 macrocells may be housed in a 388-pin package or a 272-pin packages or 100 pins. As a result of the pin configurations, the first and second packaged devices are swappable by the customer one at a time on a system printed circuit board.
In one design process, a customer designs an application (e.g., an Internet traffic controller) with a design entry software. The customer then uses a place-and-route design software to select (1) a device density that fits the application density, and (2) a package configuration based on a number of IOs suitable for the design. In the future, if the customer needs to add or remove logic, the customer can switch to a higher or lower density device, and still use the same package configuration. This may be called plug-in replacement.
Various types of memories, MUXes, DEMUXes, input/output devices, caches, controllers, registers and/or processing components may be used in accordance with the present invention. The scope of the present invention is not limited to a particular type of memory, input/output device, cache, controller, register and/or processing component. Various embodiments of the device 100 may comprise other components in addition to or instead of the components shown in the Figures without departing from the scope of the invention. For example, the device 100 in
The above-described embodiments of the present invention are merely meant to be illustrative and not limiting. It will thus be obvious to those skilled in the art that various changes and modifications may be made without departing from this invention in its broader aspects. The appended claims encompass all such changes and modifications as fall within the true spirit and scope of this invention.
Singh, Satwant, Shen, Ju, Chang, Jason
Patent | Priority | Assignee | Title |
7176716, | Dec 24 2003 | YAKIMISHU CO LTD , LLC | Look-up table structure with embedded carry logic |
7239175, | Dec 24 2003 | CALLAHAN CELLULAR L L C ; YAKIMISHU CO LTD , LLC | Look-up table based logic macro-cells |
7285984, | Dec 24 2003 | CALLAHAN CELLULAR L L C ; YAKIMISHU CO LTD , LLC | Look-up table structure with embedded carry logic |
7336097, | Dec 24 2003 | CALLAHAN CELLULAR L L C ; YAKIMISHU CO LTD , LLC | Look-up table structure with embedded carry logic |
7707237, | Aug 01 2008 | Infineon Technologies AG | Macrocell and method for adding |
9495492, | Jan 05 2015 | Cadence Design Systems, Inc. | Implementing synchronous triggers for waveform capture in an FPGA prototyping system |
Patent | Priority | Assignee | Title |
4769562, | Aug 04 1986 | CSELT - Centro Studi E Laboratori Telecommuniazioni S.p.A. | Dynamic programmable logic arrays with NOR-NOR structure implemented in C-MOS technology |
5898602, | Jan 25 1996 | XILINX, Inc.; Xilinx, Inc | Carry chain circuit with flexible carry function for implementing arithmetic and logical functions |
5977791, | Apr 15 1996 | Altera Corporation | Embedded memory block with FIFO mode for programmable logic device |
5978826, | Dec 01 1995 | AVAGO TECHNOLOGIES GENERAL IP SINGAPORE PTE LTD | Adder with even/odd 1-bit adder cells |
6020759, | Mar 21 1997 | Altera Corporation | Programmable logic array device with random access memory configurable as product terms |
6029236, | Jan 27 1997 | Altera Corporation | Field programmable gate array with high speed SRAM based configurable function block configurable as high performance logic or block of SRAM |
6064599, | Sep 03 1991 | Altera Corporation | Programmable logic array integrated circuits |
6144573, | Jun 26 1998 | Altera Corporation | Programmable logic devices with improved content addressable memory capabilities |
6150838, | Feb 25 1999 | XILINX, Inc.; Xilinx, Inc | FPGA configurable logic block with multi-purpose logic/memory circuit |
6188240, | Jun 04 1998 | HIRANUMA, TAKEO | Programmable function block |
6208163, | Feb 25 1999 | XILINX, Inc.; Xilinx, Inc | FPGA configurable logic block with multi-purpose logic/memory circuit |
6215326, | Nov 18 1998 | Altera Corporation | Programmable logic device architecture with super-regions having logic regions and a memory region |
6292017, | Mar 04 1998 | Altera Corporation | Programmable logic device incorporating function blocks operable as wide-shallow RAM |
6505226, | Sep 06 1996 | MagnaChip Semiconductor, Ltd | High speed parallel adder |
EP866558, |
Executed on | Assignor | Assignee | Conveyance | Frame | Reel | Doc |
Apr 23 2002 | SHEN, JU | Lattice Semiconductor Corporation | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 016070 | /0652 | |
Apr 23 2002 | CHAN, ALBERT | Lattice Semiconductor Corporation | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 016070 | /0652 | |
Apr 23 2002 | SINGH, SATWANT | Lattice Semiconductor Corporation | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 016070 | /0652 | |
Apr 23 2002 | CHENG, JASON | Lattice Semiconductor Corporation | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 016070 | /0652 | |
Apr 24 2002 | TSUI, CYRUS | Lattice Semiconductor Corporation | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 016070 | /0652 | |
Apr 25 2002 | LEE, CLEMENT | Lattice Semiconductor Corporation | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 016070 | /0652 | |
Feb 13 2003 | Lattice Semiconductor Corporation | (assignment on the face of the patent) | / | |||
Mar 10 2015 | SIBEAM, INC | JEFFERIES FINANCE LLC | SECURITY INTEREST SEE DOCUMENT FOR DETAILS | 035308 | /0428 | |
Mar 10 2015 | Lattice Semiconductor Corporation | JEFFERIES FINANCE LLC | SECURITY INTEREST SEE DOCUMENT FOR DETAILS | 035308 | /0428 | |
Mar 10 2015 | DVDO, INC | JEFFERIES FINANCE LLC | SECURITY INTEREST SEE DOCUMENT FOR DETAILS | 035308 | /0428 | |
Mar 10 2015 | Silicon Image, Inc | JEFFERIES FINANCE LLC | SECURITY INTEREST SEE DOCUMENT FOR DETAILS | 035308 | /0428 | |
May 17 2019 | JEFFERIES FINANCE LLC | DVDO, INC | RELEASE BY SECURED PARTY SEE DOCUMENT FOR DETAILS | 049827 | /0326 | |
May 17 2019 | JEFFERIES FINANCE LLC | SIBEAM, INC | RELEASE BY SECURED PARTY SEE DOCUMENT FOR DETAILS | 049827 | /0326 | |
May 17 2019 | JEFFERIES FINANCE LLC | Silicon Image, Inc | RELEASE BY SECURED PARTY SEE DOCUMENT FOR DETAILS | 049827 | /0326 | |
May 17 2019 | JEFFERIES FINANCE LLC | Lattice Semiconductor Corporation | RELEASE BY SECURED PARTY SEE DOCUMENT FOR DETAILS | 049827 | /0326 | |
May 17 2019 | Lattice Semiconductor Corporation | WELLS FARGO BANK, NATIONAL ASSOCIATION, AS ADMINISTRATIVE AGENT | SECURITY INTEREST SEE DOCUMENT FOR DETAILS | 049980 | /0786 |
Date | Maintenance Fee Events |
Mar 10 2005 | ASPN: Payor Number Assigned. |
Dec 04 2008 | M1551: Payment of Maintenance Fee, 4th Year, Large Entity. |
Dec 05 2012 | M1552: Payment of Maintenance Fee, 8th Year, Large Entity. |
Dec 22 2016 | M1553: Payment of Maintenance Fee, 12th Year, Large Entity. |
Date | Maintenance Schedule |
Jul 05 2008 | 4 years fee payment window open |
Jan 05 2009 | 6 months grace period start (w surcharge) |
Jul 05 2009 | patent expiry (for year 4) |
Jul 05 2011 | 2 years to revive unintentionally abandoned end. (for year 4) |
Jul 05 2012 | 8 years fee payment window open |
Jan 05 2013 | 6 months grace period start (w surcharge) |
Jul 05 2013 | patent expiry (for year 8) |
Jul 05 2015 | 2 years to revive unintentionally abandoned end. (for year 8) |
Jul 05 2016 | 12 years fee payment window open |
Jan 05 2017 | 6 months grace period start (w surcharge) |
Jul 05 2017 | patent expiry (for year 12) |
Jul 05 2019 | 2 years to revive unintentionally abandoned end. (for year 12) |